Photometric redshifts (hereafter
)
were computed using the standard
fitting procedure hyperz (Bolzonella et al. 2000) which
compares the observed spectral energy distribution (SED) of a
given galaxy, obtained from photometry, to a set of template spectra. Redshifts
are then computed using a standard
minimization. hyperz
explores the parameter space defined by the age and metallicity
of the stellar population, the IMF, the reddening law and the reddening value.
When tested against the HDF spectroscopic sample
errors
from hyperz were found to be
at
,
and
for more distant galaxies (see Bolzonella et al. 2000
for more details).
For this work, we used a set of 8 template families from the new Bruzual &
Charlot evolutionary code (GISSEL98; Bruzual & Charlot 1993) with a Miller &
Scalo IMF. The families spanned a wide range of ages for the stellar population
and included: a single burst (coeval stellar population), a constant
star-forming rate, and six -models (exponentially decaying SFR) designed to
match the sequence of colours from E-S0 to Sd galaxies (255 spectra in all). The
reddening law was taken from Calzetti et al. (2000) with values of AV
between 0 and 1.2 mag, the upper value being twice the mean E(B-V)
reported by Steidel et al. (1999) for galaxies up to
.
The Lyman
forest blanketing was modelled according to Madau (1995).
hyperz computes error bars
corresponding to 69, 90 and 99 per cent confidence levels computed by means of
the
increment for a single parameter (Avni 1976). We only
considered a
estimate when the best fit template had
1. It
may be noted that errors in the photometry of the MS 1008-1224 field were more
significant than uncertainties in the template spectra used.
The accuracy and robustness of
were investigated using simulated
catalogues of galaxies with realitics SEDs. The error budget and
accuracy were then analysed as a function of the ESO-BVRIJK filter set, the
photometric errors and the redshift range of the simulated galaxies.
First, catalogues were produced for the two sets of filters, BVRI (FORS1 images
alone) and BVRIJK (field common to FORS1 and ISAAC images), assuming a uniform
redshift distribution and a Gaussian photometric error distribution of fixed
sigma (0.1 mag), uncorrelated between the different filters.
These were used to determine (i)
errors, (ii)
the fraction of sources for which hyperz returned either no solution
(
)
or multiple solutions and (iii) the fraction of sources with
spurious values (i.e., errors much larger than the normal dispersion at
that redshift). The uniform distribution of simulated galaxies across the
redshift range provided a sufficient number of objects for a robust estimation
of the errors at all redshifts.
We then performed a second set of simulations using a pure
luminosity evolution (PLE) model. The redshift distribution and the photometric
errors (a function of magnitude and filter) in this second simulation were
tailored to mimic the VLT observations of the MS 1008-1224 field in a
more realistic manner. In particular, we focussed on (simulated) galaxies in the
range
(shear analysis sample) and
(depletion analysis sample). Galaxies were assigned magnitudes and colours
randomly according to the PLE model of Pozzetti et al. (1998) which had been
designed to reproduce the deep B counts (Williams et al. 1996).
The results from the simulations are shown in Fig. 2 (uniform redshift
distribution) and Fig. 3 (PLE distribution).
The redshift (
)
distribution of sources in the ISAAC field is shown
in Fig. 4.
The cluster of galaxies comprising MS 1008-1224 is an "in-situ'' control
sample for checking the accuracy of our
estimation. Indeed, the
cluster shows up as a prominent spike in the
distribution
(Fig. 4) between z = 0.25 and 0.4 (
)
which
confirms both the efficacy of the method as well as our error estimates from
simulations. An additional, and unexpected, check was provided by the discovery
of a background cluster at
.
That a number of galaxies were
clustered in redshift space as well as on the sky suggested that their
value was reasonably accurate.
We also simulated cluster fields at z = 0.31 as targets for the hyperz
program. The clusters were generated with galaxies distributed according to a
King model with central line-of-sight velocity dispersion of 1000 kms-1,
a core radius of 500 kpc and a Schecter luminosity distribution in the range
.
The mixture included 70 per cent ellipticals and S0
galaxies, 28 per cent spirals and 2 per cent star-forming galaxies. The other
parameters (IMF, SFR, models etc.) were as described before. The photometric
accuracy (as a function of magnitude) and limiting magnitudes were chosen to
match the observed values for MS 1008-1224. The apparent magnitudes in all
filters were computed through GISSEL98. This simulated cluster catalogue was
added to a PLE field catalogue to simulate the observed catalogue.
Figure 5 shows the cluster sequence on the observed Colour-Magnitude
plot R-I vs. R.
Simulations indicated that the error in
for cluster members was
0.04 (
)
which is similar to the width of the peak obtained for
real data (Fig. 4). It may be noted that this accuracy is much
better than
because of the presence of
appropriately located spectral features which make identification of red cluster
galaxies particularly easy in this redshift range. It is also clear that the
cluster redshift distribution is skewed, the lower redshift side of the peak
being considerably steeper than its counterpart. So we defined the foreground
galaxy sample as those at
and the background sample (for
lensing analysis) as those at (conservatively)
.
hyperz
detected 75 per cent of the simulated cluster galaxies in the range
with BVRIJK.
One of the problems with using the BVRI photometry for
is the
contamination of lower redshift bins by interlopers from z > 1 and this could
be higher than 1 in 3 sources. On the contrary, the simulated Colour-Magnitude
diagram in Fig. 4 indicated that the cluster luminosity is
dominated (80-90 per cent) by emission from red ellipticals on the cluster
sequence. Carlberg et al. (1996) estimated that the red galaxies on the
cluster sequence underestimate the cluster luminosity by about 15 per cent,
similar to what we see in our simulated data. So we used galaxies from the
entire FORS1 field on the cluster sequence of the Colour-Magnitude diagram
(
,
17.5 < R < 24.0) for calculating the luminosity
distribution and applied the
correction determined above.
The number and luminosity density distributions of cluster galaxies
are shown in Fig. 6.
![]() |
Figure 6:
Galaxy number and R-band luminosity density
distribution in the galaxy cluster MS 1008-1224. The densities were computed
from galaxies on the cluster sequence on the Colour-Magnitude plot (see Fig. 5 and the text) and smoothed with a 50'' Gaussian. The average
galaxy number density is 5.9 gal arcmin-1. The
number density contours plotted (left) are 12.0, 24.0, 36.0, 72.0, 144.0, 215.0,
280.0, 360.0, 720.0, 1080.0, 1440.0, and 1800.0 gal arcmin-2. So, the
density contrast reaches 305 in the cluster center with respect to the average
value. The average luminosity over the field is
![]() ![]() |
We used photometric redshifts to calculate the average lensing distance
modulus to convert the gravitational convergence (shear analysis) into a mass
estimate. Thus missing the objects from BVRI photometry would have
resulted in a considerable error in the absolute mass estimate. Therefore we
assumed that the redshift distribution in the ISAAC field (i.e. those sources
which had BVRIJK magnitudes) was representative of the whole FORS1 field.
The redshift of background sources affects the mass estimate only through the
angular scale distance which has a weak redshift dependence at z > 0.5 in
most cosmologies. So
is not of much concern,
especially when compared to other sources of
error discussed in the section on shear analysis. We also used the
distribution to determine the foreground source contamination of the lensing
sample which would have diluted the lensing signal and corrected for the same.
Copyright ESO 2002